Read blurry receipt photos and handwriting
untxt. reads low-quality input, blurry phone photos, dim scans, crumpled receipts, and extracts what is legible. Where a value is too unclear to be sure, it flags it for review instead of guessing. Clear printed content is read reliably; heavy handwriting is the honest limit.
Real receipts are not clean scans
Clients photograph receipts on a table, in bad light, half-crumpled, sometimes faded to grey. That is the normal input, not a worst case, and untxt. is built to read it rather than bounce it back.
What it reads reliably, and what it flags
Clear printed text comes through dependably, even from a poor photo. As legibility drops, untxt. extracts what it can read and attaches a confidence score, so a value it is unsure about is surfaced for you instead of entered wrong.
Handwriting is the honest limit
Here is the straight answer: printed content is where untxt. is strongest, and handwriting is the hardest case in document reading, for any tool. untxt. reads clear handwriting where it can and flags the rest for you to confirm. It tells you when it is unsure rather than inventing a number to fill the field.
How a low-confidence read shows up
In the review queue every document carries a confidence score, and the original sits beside the extracted data. A blurry total is one glance to check and one click to fix, so your attention goes to the few unclear values, not the clear ones you would otherwise re-key.